Autonomous harvest and transportation is a long-time goal of the forest industry. Compared with the urban environment, the natural forest is a large unstructured environment where less human-made features like straight edge and the flat plane can be observed from sensors. Hence some state of the art method developed for urban environments like perception and path planning can not be directly utilized in the forest environment. This project leverage the current most advanced sensors, through multi-sensor fusion and deep learning, to develop a more efficient and accurate localization, mapping and navigation algorithm for the autonomous robot in the forest environment.